JPX Tokyo Stock Exchange Prediction: Tokyo Gas Study Case

This project was built by a team of four persons including Azizha Zeinita

Japan Exchange Group, Inc. (JPX) is a holding company operating one of the largest stock exchanges in the world, which one of it’s significant stocks is Tokyo Gas stock. This project will compare several models against real future returns& of Tokyo Gas stock after the training phase is complete. Other external datasets that are closely related to stock price, such as Japan GDP, Unemployment Rate, Oil Price, etc., will also involved in different combinations to find the best output.

Time series models using ARIMA, VAR, PROPHET, Long Short-Term Memory (LSTM) with evaluation metrics MAE, MSE, MAPE, sMAPE, MASE

Preliminaries

Functions

Stock Data Load

External Data Load

EDA

Industry EDA

Energy Comapny list

Merge data

Plots by category

Elextric Industry

Gas Industry

Oil Industry

Main 5 Companies

ARIMA Models(Tokyo Gas)

Daily Analysis ARIMA(1 Year Prediction)

Univariate ARIMA

Multivariate ARIMAX

With Oil Regressor

With NIKKEI225 Regressor

Oil & Nikkei

Weekly Analysis ARIMA (6 Monthes Prediction)

Pre-processing & stationarity check for weekly data

Pre-processing for external regressors

Data Splitting

ARIMA Model Pipeline and Visualization

Univariate ARIMA

Multivariate ARIMAX

NIKKEI 225 Regressor

Oil Regressor

All external regressors

VAR Model (Tokyo Gas)

Pre-processing & stationarity check

Vector Auto Regression (VAR) - Oil

Vector Auto Regression (VAR) - Oil & N225

Vector Auto Regression (VAR) - Oil, N225, Unemployment

Vector Auto Regression (VAR) - Oil, N225, Unemployement, GDP

Vector Auto Regression (VAR) - Oil, N225, Unemployement, GDP, USD-JPY rate

Vector Auto Regression (VAR) - All External Data

Prophet (Tokyo Gas)

Daily Analysis Prophet (1 Year Prediction)

Univariate Prophet

Multivariate Prophet

N225 & Oil

Weekly Analysis Prophet (6 Monthes Prediction)

Univariate Prophet

Multivariate Prophet

All External regressors
only n225
oil & usdjpy
GDP & Unemp

Tunnned Daily Prophet 1 Year Prediction

Hyperparameter Tunnning

Residuals Check for Prohpet

Prediction and Accuracy

LSTM Model

RNN

LSTM

Recursive model (ENEOS)

Rolling model only applied to the stock ENEOS because the runtime was too long when rolling model run on other stocks we picked

Pre-processing

Pre-processing & stationarity check

Pre-processing for external regressors

Data Splitting

Function

Rolling Univariate ARIMA (3)

Rolling Univariate ARIMA (4)

All External regressors

only n225

oil & usdjpy

GDP & Unemp

Rolling ARIMAX with External Regressors (4)

All External regressors

only n225

oil & usdjpy

GDP & Unemp

Rolling VAR

All external regressors

only n225

oil & usdjpy

GDP & Unemp

Rolling VAR2

All external regressors

only n225

oil & usdjpy

GDP & Unemp

Summarize results for rolling fit models

Summarize result for daily data

Summarize result for weekly analysis